To develop a model for classifying normal controls (NC), mild cognitive impairment (MCI), and Alzheimer's disease (AD) using structural MRI, addressing challenges in anatomical variations and disease progression.
Approach:
Model Architecture: APLG-Net integrates a global whole-brain encoder and a local ROI-based encoder, utilizing cross-attention fusion and vector-gated integration.
Supervision Strategy: An ordinal supervision strategy is introduced to model disease progression among NC, MCI, and AD.
Key Findings:
APLG-Net achieved 87.1% accuracy, 86.4% balanced accuracy, 86.8% Macro-F1, and 85.6% MCI F1 on the ADNI dataset.
The model outperformed CNN-based, Transformer-based, and hybrid baselines.
APLG-Net effectively addresses the challenges of classifying NC, MCI, and AD by integrating local and global anatomical information and modeling disease progression.
